Improve thread safety of SwaggerClient

[mckinsel]

SwaggerClient discovers methods by reading a swagger document, creating
instances of _ClientMethodFactory for paths in the swagger, and
assigning those instances as class attributes of SwaggerClient.

This means that a single instance of each _ClientMethodFactory is shared
across threads. This is not so good because _ClientMethodFactories
store state associated with requests. In particular, consider the
following scenario:

Thread-1: client = hca.dss.DSSClient()
Thread-2: client = hca.dss.DSSClient()
Thread-1: with client.get_file.stream(...) as handle:
Thread-2: with client.get_file.stream(...) as handle:
Thread-1: handle.raw.read()

Depending on the execution of __enter__, Thread-1 may have just read from
the file for Thread-2! Going on,

Thread-1: __exit__ the with block
Thread-2: handle.raw.read()
Thread-2: __exit__ the with block
Exception: NoneType has no attribute close()

Or you know, something like that.

This change makes the methods instance attributes.

[ttung]

unit tests are broken, btw.

my suggestion for making this a bit safer: in __init__(..), create a lock. in stream(..), hold the lock, and verify that _context_manager_response is None before issuing the request. if it's not None, then the previous request is plausibly still running.

[mckinsel]

That's definitely safer, but I think the safety is achieved by fully serializing calls to stream. Once one thread calls stream, no other thread could until that first thread calls __exit__.

[mckinsel]

Also, re unit tests, I'm pretty sure master is broken? Is that possible?

[ttung]

That's definitely safer, but I think the safety is achieved by fully serializing calls to stream. Once one thread calls stream, no other thread could until that first thread calls exit.

I think my suggestion achieves that same effect...?

[ttung]

Also, re unit tests, I'm pretty sure master is broken? Is that possible?

Yes, that's definitely possible.

[kozbo]

Passing now, would you like to try your unit tests again?

[mckinsel]

Thanks @kozbo , still failing, but interesting failures now.

[kozbo]

@mckinsel any progress on this? Is this still an issue?